Machine Learning Docker Images for video research

This repo contains a set of Docker Images which support

FFMPEG built with NVEnc and additional GPU codecs enabled (HAP)
Build on top of standard ML Base docker images, such as Pytorch GPU from Nvidia and ONNX GPU from Microsoft.

Please see the Dockerfiles at :

FFMPEG 4.1 - Pytorch GPU

FFMPEG 4.1 - ONNX GPU

Notice

This Repo contains a fork of @jrottenberg's amazing FFMPEG docker repository.

Changes include:

Add HAP support for FFMPEG 4.1 Nvidia container
Ensure FFMPEG 4.1 properly supports MXF / OP1/A Motion JPEG 2000 codec (4.2, 4.3 and newer appear to have decode issues)
Keep NVidia CUDA / Dev libs available for downstream docker containers to leverage.

The intent of this repository is to provide a baseline Docker useful for:

Hardware Acceelerated video processing using FFMPEG
GPU accelerated post processing
Base image for Machine Learning video workflows that want HWAccel support and variant codecs.

Please note only FFMPEG 4.1 / NVidia is supported until further notice.

Original Documentation:

FFmpeg Docker image

This project prepares a minimalist Docker image with FFmpeg. It compiles FFmpeg from sources following instructions from the Compilation Guide.

You can install the latest build of this image by running docker pull jrottenberg/ffmpeg.

This image can be used as a base for an encoding farm.

Builds

There are different builds available:

alpine based images ffmpeg:<version>-alpine38 or ffmpeg:<version>-alpine312 (the ffmpeg:<version>-alpine builds are not updated any more)
- alpine based scratch images ffmpeg:<version>-scratch38 or ffmpeg:<version>-scratch312 (experimental image containing only FFmpeg and libraries, the ffmpeg:<version>-alpine builds are not updated any more)
centos based images ffmpeg:<version>-centos7 or ffmpeg:<version>-centos8
ubuntu based images ffmpeg:<version>-ubuntu1804 or ffmpeg:<version>-ubuntu2004 (the default, you can also use ffmpeg:<version>-ubuntu as an alias)
- ubuntu based nvidia images ffmpeg:<version>-nvidia1804
- ubuntu based vaapi images ffmpeg:<version>-vaapi1804 or ffmpeg:<version>-vaapi2004

<version> can be one of the following:

3.2
3.3
3.4
4.0
4.1
4.2
4.3
snapshot

Not all combinations are supported and older versions will fade out over time. See the table below for the currently supported combinations.

Version	alpine38	alpine312	centos7	centos8	nvidia1804	scratch38	scratch312	ubuntu1804	ubuntu2004	vaapi1804	vaapi2004
3.2	✔️	❌	✔️	❌	❌	✔️	❌	❌	✔️	❌	✔️
3.3	❌	✔️	✔️	✔️	✔️	❌	✔️	✔️	✔️	❌	✔️
3.4	❌	✔️	✔️	✔️	✔️	❌	✔️	✔️	✔️	❌	✔️
4.0	❌	✔️	✔️	✔️	✔️	❌	✔️	✔️	✔️	❌	✔️
4.1	❌	✔️	✔️	✔️	✔️	❌	✔️	✔️	✔️	❌	✔️
4.2	❌	✔️	✔️	✔️	✔️	❌	✔️	✔️	✔️	❌	✔️
4.3	✔️	✔️	✔️	✔️	✔️	✔️	✔️	✔️	✔️	✔️	✔️
snapshot	❌	✔️	✔️	✔️	✔️	❌	✔️	✔️	✔️	❌	✔️

Generate list of recent images

You can use the following command to generate a list of current images:

$ curl --silent https://hub.docker.com/v2/repositories/jrottenberg/ffmpeg/tags/?page_size=500 | jq -cr ".results|sort_by(.name)|reverse[]|.sz=(.full_size/1048576|floor|tostring+\"mb\")|[.name,( (20-(.name|length))*\" \" ),.sz,( (8-(.sz|length))*\" \"),.last_updated[:10]]|@text|gsub(\"[,\\\"\\\]\\\[]\";null)"

If you want to compare the one you have locally, use the following command:

$ docker images | grep ffmpeg | sort | awk '{print $1 ":" $2 "\t" $7 $8}'

Please use Github issues to report any bug or missing feature.

Test

ffmpeg version N-98740-ga72d529 Copyright (c) 2000-2020 the FFmpeg developers
  built with gcc 7 (Ubuntu 7.5.0-3ubuntu1~18.04)
  configuration: --disable-debug --disable-doc --disable-ffplay --enable-shared --enable-avresample --enable-libopencore-amrnb --enable-libopencore-amrwb --enable-gpl --enable-libass --enable-fontconfig --enable-libfreetype --enable-libvidstab --enable-libmp3lame --enable-libopus --enable-libtheora --enable-libvorbis --enable-libvpx --enable-libwebp --enable-libxcb --enable-libx265 --enable-libxvid --enable-libx264 --enable-nonfree --enable-openssl --enable-libfdk_aac --enable-postproc --enable-small --enable-version3 --enable-libbluray --enable-libzmq --extra-libs=-ldl --prefix=/opt/ffmpeg --enable-libopenjpeg --enable-libkvazaar --enable-libaom --extra-libs=-lpthread --enable-libsrt --enable-libaribb24 --enable-vaapi --extra-cflags=-I/opt/ffmpeg/include --extra-ldflags=-L/opt/ffmpeg/lib
  libavutil      56. 58.100 / 56. 58.100
  libavcodec     58.100.100 / 58.100.100
  libavformat    58. 51.100 / 58. 51.100
  libavdevice    58. 11.101 / 58. 11.101
  libavfilter     7. 87.100 /  7. 87.100
  libavresample   4.  0.  0 /  4.  0.  0
  libswscale      5.  8.100 /  5.  8.100
  libswresample   3.  8.100 /  3.  8.100
  libpostproc    55.  8.100 / 55.  8.100

  configuration:
    --disable-debug
    --disable-doc
    --disable-ffplay
    --enable-shared
    --enable-avresample
    --enable-libopencore-amrnb
    --enable-libopencore-amrwb
    --enable-gpl
    --enable-libass
    --enable-fontconfig
    --enable-libfreetype
    --enable-libvidstab
    --enable-libmp3lame
    --enable-libopus
    --enable-libtheora
    --enable-libvorbis
    --enable-libvpx
    --enable-libwebp
    --enable-libxcb
    --enable-libx265
    --enable-libxvid
    --enable-libx264
    --enable-nonfree
    --enable-openssl
    --enable-libfdk_aac
    --enable-postproc
    --enable-small
    --enable-version3
    --enable-libbluray
    --enable-libzmq
    --extra-libs=-ldl
    --prefix=/opt/ffmpeg
    --enable-libopenjpeg
    --enable-libkvazaar
    --enable-libaom
    --extra-libs=-lpthread
    --enable-libsrt
    --enable-libaribb24
    --enable-vaapi
    --extra-cflags=-I/opt/ffmpeg/include
    --extra-ldflags=-L/opt/ffmpeg/lib

Capture output from the container to the host running the command

 docker run jrottenberg/ffmpeg \
            -i http://url/to/media.mp4 \
            -stats \
            $ffmpeg_options  - > out.mp4

Examples

Extract 5s @00:49:42 into a GIF

 docker run jrottenberg/ffmpeg -stats  \
        -i http://archive.org/download/thethreeagesbusterkeaton/Buster.Keaton.The.Three.Ages.ogv \
        -loop 0  \
        -final_delay 500 -c:v gif -f gif -ss 00:49:42 -t 5 - > trow_ball.gif

Convert 10bits MKV into a 10Bits MP4

 docker run -v $(pwd):$(pwd) -w $(pwd) jrottenberg/ffmpeg:3.4-scratch \
        -stats \
        -i http://www.jell.yfish.us/media/jellyfish-20-mbps-hd-hevc-10bit.mkv \
        -c:v libx265 -pix_fmt yuv420p10 \
        -t 5 -f mp4 test.mp4

The image has been compiled with X265 Multilib. Use the pixel format switch to change the number of bits per pixel by suffixing it with 10 for 10bits or 12 for 12bits.

Convert a local GIF into a mp4

Let's assume original.gif is located in the current directory :

 docker run -v $(pwd):$(pwd) -w $(pwd)\
        jrottenberg/ffmpeg:3.2-scratch -stats \
        -i original.gif \
        original-converted.mp4

Use ZeroMQ to toggle filter value on-fly

Let's start some process continuously writing some radio music, and listen it:

 docker run --rm -d -v $(pwd):$(pwd) -w $(pwd) -p 11235:11235 \
        --name radio-writer jrottenberg/ffmpeg \
        -i http://radio.casse-tete.solutions/salut-radio-64.mp3 \
        -filter_complex '[0:a]volume@vol=1,azmq=bind_address=tcp\\\://0.0.0.0\\\:11235[out]' \
        -map '[out]' ./salut-radio.mp3

 ffplay ./salut-radio.mp3

Now, just toggle its volume on-fly, and hear how it changes:

 docker run --rm --network=host --entrypoint sh jrottenberg/ffmpeg -c \
        'echo "volume@vol volume 2" | zmqsend -b tcp://127.0.0.1:11235'

Send a stream over SRT

Let's send video.mp4 to srt-listener on port 9000 over SRT protocol.

docker run -v $(pwd):$(pwd) jrottenberg/ffmpeg \
       -re -i $(pwd)/video.mp4 -acodec copy -vcodec copy -f mpegts srt://srt-listener:9000?pkt_size=1316

Use hardware acceleration enabled build

Thanks to qmfrederik for the vaapi ubuntu based variant

jrottenberg/ffmpeg:vaapi or jrottenberg/ffmpeg:${VERSION}-vaapi

Run the container with the device attached /dev/dri from your host into the container :

docker run --device /dev/dri:/dev/dri -v $(pwd):$(pwd) -w $(pwd) jrottenberg/ffmpeg:vaapi [...]

Have the Intel drivers up and running on your host. You can run vainfo (part of vainfo package on Ubuntu) to determine whether your graphics card has been recognized correctly.
Run ffmpeg with the correct parameters, this is the same as when running ffmpeg natively.

Use nvidia hardware acceleration enabled build

Thanks to ShaulMyplay for the nvidia based variant

Supports nvenc only on all ffmpeg versions, and hardware decoding and scaling on ffmpeg >= 4.0

Install nvidia latest drivers on host machine.
Install nvidia-docker on host machine.
Run container using "--runtime=nvidia" flag and use supported ffmpeg hwaccel options

Hardware encoding only example:

docker run --runtime=nvidia jrottenberg/ffmpeg:2.8-nvidia -i INPUT -c:v nvenc_h264 -preset hq OUTPUT Full hardware acceleration example: docker run --runtime=nvidia jrottenberg/ffmpeg:4.1-nvidia -hwaccel cuvid -c:v h264_cuvid -i INPUT -vf scale_npp=-1:720 -c:v h264_nvenc -preset slow OUTPUT

See what's inside the beast

docker run -it --entrypoint='bash' jrottenberg/ffmpeg

for i in ogg amr vorbis theora mp3lame opus vpx xvid fdk x264 x265;do echo $i; find /usr/local/ -name *$i*;done

Keep up to date

See Dockerfile-env to update a version

Contribute

# Add / fix stuff
${EDITOR} templates/

# Generates the Dockerfile for all variants
./update.py

# Test a specific variant
docker build -t my-build docker-images/VERSION/

# Make sure all variants pass before CI
find ffmpeg/ -name Dockerfile | xargs dirname | parallel --no-notice -j 4 --results logs docker build -t {} {}

Commit the templates files THEN all the generated Dockerfile for a merge request. So it's easier to review the template change.

Name		Name	Last commit message	Last commit date
Latest commit History 332 Commits
docker-images		docker-images
templates		templates
.gitlab-ci.yml		.gitlab-ci.yml
LICENSE		LICENSE
README.md		README.md
azure-pipelines.yml		azure-pipelines.yml
generate-stackbrew-library.sh		generate-stackbrew-library.sh
requirements-dev.txt		requirements-dev.txt
setup.cfg		setup.cfg
update.py		update.py

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Repository files navigation

Machine Learning Docker Images for video research

Notice

Changes include:

Original Documentation:

FFmpeg Docker image

Builds

Generate list of recent images

Test

Examples

Extract 5s @00:49:42 into a GIF

Convert 10bits MKV into a 10Bits MP4

Convert a local GIF into a mp4

Use ZeroMQ to toggle filter value on-fly

Send a stream over SRT

Use hardware acceleration enabled build

Use nvidia hardware acceleration enabled build

See what's inside the beast

Keep up to date

Contribute

About

Languages

License

Synopsis/ml-ffmpeg-docker

Folders and files

Latest commit

History

Repository files navigation

Machine Learning Docker Images for video research

Notice

Changes include:

Original Documentation:

FFmpeg Docker image

Builds

Generate list of recent images

Test

Examples

Extract 5s @00:49:42 into a GIF

Convert 10bits MKV into a 10Bits MP4

Convert a local GIF into a mp4

Use ZeroMQ to toggle filter value on-fly

Send a stream over SRT

Use hardware acceleration enabled build

Use nvidia hardware acceleration enabled build

See what's inside the beast

Keep up to date

Contribute

About

Topics

Resources

License

Stars

Watchers

Forks

Languages